After a fair bit of experimentation I've come to the conclusion that for me at least, AI is not yet quite *there* in terms of generating photorealistic images, but it is very good at creating more art-type images, like drawings or paintings, which follow more in the tradition of WAM stories than photography.

So here are some 21st century pre-raphaelite paintings featuring three of the women of a university rowing club, relaxing in the deep mudbaths of a castle's bathing chamber after a race, while still wearing their sporty spandex rowing catsuits and wellies.

All made with https://promptus.ai

I'd probably caveat that and say the right model with the right prompts can most certainly produce near indistinguishable images. But... Not every possible scene and complexity. I'd say that is regardless of the output style, line drawing, art or otherwise.

One other thing.. Promptus isn't the AI. Id say which model produced it whether it's Sora, GPT4o, Imagen4, Nanobanana, FLux etc.

Near, yes, but not yet there, and I suspect getting there will take a while yet. Bear in mind I both generated AI images and shoot real-world wam with real models. Ok, what AI has achived over the last two years has been impressive - remember when every second image had people with seventeen fingers and a second set of teeth inside their outer ones? But that "last mile" to actually simulating genuine humanity - that's still a long way off. Doesn't mean it won't happen, but not there yet. What I see through my cameras when I shoot real wam is an multiple entire orders of magnitude away from even the very best AI.


Promptus is the system I used to do it, models mostly Flux and NanoBanana, though I've played with others. SeeDream produces impressive painting-style results but isn't that good at faces. Weirdly whenever I've used a GPT model the result has been very dark compared to any others, not sure why, unless that's something it's interpreting in the pre-raphaelite suggestion parts of the prompt.

I think this is always going to be the case. True real photography and video conveys emotion with context. I'd say I've generated some truely close lifelike images but digging deep there is always something that will throw it off if you look deep enough. a quiver in the lip, a wrinkle not there in the brow, a frayed stitch not quite right, some distortion in the background etc. I doubt it'll ever be perfect and really, I don't think it needs to be and also why I think producers have little to fear.

That said, you have the same issues with paintings, or other art styles. the brush strokes don't look real or the raised pain of an oil painting.


Each model has it's own nuances in prompting input needed. You can't just lift a prompt from one model to another. To get the best from the models each prompt needs to be specifically tailored. GPT has a "colour" which is blatently obvious in 95% of the images I see. It can be prompted out with system prompts and the main prompt but it has issues.

Ultimately though, I'd say the majority of users, just don't know how to prompt and I mean that as a general observation of what people post elsewhere.. How many people tailor a system prompt for a photographic style or art style? The current AI doesn't know what you want so you don't get the result you are thinking of

I've got a range of system prompts that I used depending on what I'm trying to create for studio, candid and period shots. Each with fairly detailed instructions.

## 1. PHOTOGRAPHIC STANDARDS (NON-NEGOTIABLE)
## 2. ANATOMY & TEXTURE FIDELITY
## 3. FLUID DYNAMICS & MATERIAL PHYSICS
## 4. CLOTHING INTERACTION
## 5. GENERATION GUIDANCE

i.e.

## 1. PHOTOGRAPHIC STANDARDS (NON-NEGOTIABLE)
- **Camera System:** Phase One XF IQ4 150MP Trichromatic.
- **Lens Optics:** Simulate an 85mm or 105mm Macro Prime lens at f/2.8 to f/4.
- **Focus:** Razor-sharp focus on the eyes/lashes. Implement a natural, creamy optical bokeh fall-off towards the ears and background to simulate shallow depth of field.
- **Lighting Setup:** Professional 3-point studio setup.
- *Key:* Large Octabank or Softbox (soft, wrapping light).
- *Fill:* White reflector to lift shadows.
- *Rim/Hair:* Hard light to define edges and separate subject from the background.
- *Reflections:* Lighting must reflect realistically off wet substances (specular highlights).